It also turns out that it allows for a bunch more compiler optimizations to be implimented with a lot less pain. I got very nerd sniped on this this week leading to
https://github.com/JuliaLang/julia/pull/56030 and
https://github.com/JuliaLang/julia/pull/55913 which allow allocation removal in a number of cases and saves ~4ns (~10ns->6ns for Memory{Int8}(unef, 4)) for constructing Memory objects.